Noninvasive positive-pressure ventilation (hereafter, noninvasive ventilation) reduces the need for endotracheal intubation and mortality among patients with acute exacerbations of chronic obstructive pulmonary disease1-3 or severe cardiogenic pulmonary edema.4 The physiological effects of noninvasive ventilation include a decrease in the work of breathing and improvement in gas exchange. In patients with acute hypoxemic respiratory failure, the need for mechanical ventilation is associated with high mortality,5 but data on the overall effects of noninvasive ventilation with respect to the prevention of intubation and improvement in outcome are conflicting.6-10Previous studies have often included a heterogeneous population of patients with acute respiratory failure who had chronic lung disease7,10 or cardiogenic pulmonary edema8,9; this selection of patients could lead to an overestimation of the beneficial effects of noninvasive ventilation as compared with standard oxygen therapy. In observational studies focusing on patients with acute hypoxemic respiratory failure, the rate of treatment failure with noninvasive ventilation was as high as 50%11-13 and was often associated with particularly high mortality.14,15 To date, the literature does not conclusively support the use of noninvasive ventilation in patients with nonhypercapnic acute hypoxemic respiratory failure.High-flow oxygen therapy through a nasal cannula is a technique whereby heated and humidified oxygen is delivered to the nose at high flow rates. These high flow rates generate low levels of positive pressure in the upper airways, and the fraction of inspired oxygen (FIO2) can be adjusted by changing the fraction of oxygen in the driving gas.16-18 The high flow rates may also decrease physiological dead space by flushing expired carbon dioxide from the upper airway, a process that potentially explains the observed decrease in the work of breathing.19 In patients with acute respiratory failure of various origins, high-flow oxygen has been shown to result in better comfort and oxygenation than standard oxygen therapy delivered through a face mask.20-25To our knowledge, the effect of high-flow oxygen on intubation rate or mortality has not been assessed in patients admitted to the intensive care unit (ICU) with acute hypoxemic respiratory failure. We conducted a prospective, multicenter, randomized, controlled trial involving patients admitted to the ICU with acute hypoxemic respiratory failure to determine whether high-flow oxygen therapy or noninvasive ventilation therapy, as compared with standard oxygen therapy alone, could reduce the rate of endotracheal intubation and improve outcomes.MethodsStudy OversightWe conducted the study in 23 ICUs in France and Belgium. For all the centers in France, the study protocol (available with the full text of this article at NEJM.org) was approved by the ethics committee at Centre Hospitalier Universitaire de Poitiers; for the study site at Cliniques Universitaires Saint-Luc, Brussels, the protocol was approved by the ethics committee at that center. Written informed consent was obtained from all the patients, their next of kin, or another surrogate decision maker as appropriate.The trial was overseen by a steering committee that presented information regarding the progression and monitoring of the study at Rseau Europen de Recherche en Ventilation Artificielle (REVA) Network meetings every 4 months. An independent safety monitoring board was set up. Research assistants regularly monitored all the centers on site to check adherence to the protocol and the accuracy of the data recorded. An investigator at each center was responsible for enrolling patients in the study, ensuring adherence to the protocol, and completing the electronic case-report form. Although the individual study assignments of the patients could not be masked, the coordinating center and all the investigators remained unaware of the study-group outcomes until the data were locked in July 2014. All the analyses were performed by the study statistician, in accordance with the International Conference on Harmonisation and Good Clinical Practice guidelines. Face masks, heated humidifiers, and cannulas (i.e., consumable materials) were donated to the participating ICUs, and air-oxygen blenders were provided during the study period, by Fisher and Paykel Healthcare, which had no other involvement in the study.PatientsConsecutive patients who were 18 years of age or older were enrolled if they met all four of the following criteria: a respiratory rate of more than 25 breaths per minute, a ratio of the partial pressure of arterial oxygen (PaO2) to the FIO2 of 300 mm Hg or less while the patient was breathing oxygen at a flow rate of 10 liters per minute or more for at least 15 minutes, a partial pressure of arterial carbon dioxide (PaCO2) not higher than 45 mm Hg, and an absence of clinical history of underlying chronic respiratory failure. FIO2 was measured by a portable oxygen analyzer (MX300, Teledyne Analytical Instruments) that was introduced in the nonrebreather face mask.The main exclusion criteria were a PaCO2 of more than 45 mm Hg, exacerbation of asthma or chronic respiratory failure, cardiogenic pulmonary edema, severe neutropenia, hemodynamic instability, use of vasopressors, a Glasgow Coma Scale score of 12 points or less (on a scale from 3 to 15, with lower scores indicating reduced levels of consciousness), contraindications to noninvasive ventilation, urgent need for endotracheal intubation, a do-not-intubate order, and a decision not to participate. Details of the study exclusion criteria are provided in the Supplementary Appendix, available at NEJM.org.RandomizationRandomization was performed in permuted blocks of six, with stratification according to center and history or no history of cardiac insufficiency. Within 3 hours after the validation of inclusion criteria, patients were randomly assigned in a 1:1:1 ratio, with the use of a centralized Web-based management system (Clinsight, Ennov), to one of the three following strategies: high-flow oxygen therapy, standard oxygen therapy, or noninvasive ventilation.InterventionsIn the standard-oxygen group, oxygen therapy was applied continuously through a nonrebreather face mask at a flow rate of 10 liters per minute or more. The rate was adjusted to maintain an oxygen saturation level of 92% or more, as measured by means of pulse oximetry (Spo2), until the patient recovered or was intubated.In the high-flowoxygen group, oxygen was passed through a heated humidifier (MR850, Fisher and Paykel Healthcare) and applied continuously through large-bore binasal prongs, with a gas flow rate of 50 liters per minute and an FIO2 of 1.0 at initiation (Optiflow, Fisher and Paykel Healthcare). The fraction of oxygen in the gas flowing in the system was subsequently adjusted to maintain an Spo2 of 92% or more. High-flow oxygen was applied for at least 2 calendar days; it could then be stopped and the patient switched to standard oxygen therapy.In the noninvasive-ventilation group, noninvasive ventilation was delivered to the patient through a face mask (Fisher and Paykel Healthcare) that was connected to an ICU ventilator, with pressure support applied in a noninvasive-ventilation mode. The pressure-support level was adjusted with the aim of obtaining an expired tidal volume of 7 to 10 ml per kilogram of predicted body weight, with an initial positive end-expiratory pressure (PEEP) between 2 and 10 cm of water. The FIO2 or PEEP level (or both) were then adjusted to maintain an Spo2 of 92% or more. The minimally required duration of noninvasive ventilation was 8 hours per day for at least 2 calendar days. Noninvasive ventilation was applied during sessions of at least 1 hour and could be resumed if the respiratory rate was more than 25 breaths per minute or the Spo2 was less than 92%. Between noninvasive-ventilation sessions, patients received high-flow oxygen, as described above.Study OutcomesThe primary outcome was the proportion of patients who required endotracheal intubation within 28 days after randomization. To ensure the consistency of indications across sites and reduce the risk of delayed intubation, the following prespecified criteria for endotracheal intubation were used: hemodynamic instability, a deterioration of neurologic status, or signs of persisting or worsening respiratory failure as defined by at least two of the following criteria: a respiratory rate of more than 40 breaths per minute, a lack of improvement in signs of high respiratory-muscle workload, the development of copious tracheal secretions, acidosis with a pH of less than 7.35, an Spo2 of less than 90% for more than 5 minutes without technical dysfunction, or a poor response to oxygenation techniques (details of the criteria are provided in the Supplementary Appendix). In the high-flowoxygen group and the standard-oxygen group, a trial of noninvasive ventilation was allowed at the discretion of the physician in patients who had signs of persisting or worsening respiratory failure and no other organ dysfunction before endotracheal intubation was performed and invasive ventilation initiated.Secondary outcomes were mortality in the ICU, mortality at 90 days, the number of ventilator-free days (i.e., days alive and without invasive mechanical ventilation) between day 1 and day 28, and the duration of ICU stay. Other prespecified outcomes included complications during the ICU stay, such as septic shock, nosocomial pneumonia, cardiac arrhythmia, and cardiac arrest. Dyspnea was assessed with the use of a 5-point Likert scale, and comfort with the use of a 100-mm visual-analogue scale (see the Supplementary Appendix).Statistical AnalysisAssuming an intubation rate of 60% in the population that was treated with standard oxygen therapy,7,9,10 we calculated that enrollment of 300 patients would provide the study with 80% power to show an absolute difference of 20 percentage points in the primary outcome between the standard-oxygen group and either the high-flowoxygen group or the noninvasive-ventilation group at a two-sided alpha level of 0.05.All the analyses were performed on an intention-to-treat basis. KaplanMeier curves were plotted to assess the time from enrollment to endotracheal intubation or death and were compared by means of the log-rank test.The treatment (standard oxygen, high-flow oxygen, or noninvasive ventilation) was introduced as two dummy variables to obtain two odds ratios or hazard ratios for comparison with the reference group, which was defined as the lowest-risk group. Variables associated with intubation at day 28 and in-ICU mortality were assessed by means of multivariate logistic-regression analyses, and those associated with mortality at 90 days were assessed by means of a Cox proportional-hazard regression analysis with the use of a backward-selection procedure. The final model included a history of cardiac insufficiency and variables significantly associated with intubation or mortality with a P value of less than 0.05.We conducted only one post hoc subgroup analysis, which included patients with a PaO2:FIO2 of 200 mm Hg or less at enrollment, to analyze outcomes in patients with severe hypoxemia. This threshold of the PaO2:FIO2 was based on the classification of the acute respiratory distress syndrome.26-28A two-tailed P value of less than 0.05 was considered to indicate statistical significance. We used SAS software, version 9.2 (SAS Institute), for all the analyses.ResultsPatientsFrom February 2011 through April 2013, a total of 2506 patients with acute hypoxemic respiratory failure were admitted to the 23 participating ICUs; 525 patients were eligible for inclusion in the study, and 313 underwent randomization (Figure 1Figure 1Enrollment, Randomization, and Follow-up of the Study Participants.). After the secondary exclusion of 3 patients who withdrew consent, 310 patients were included in the analysis. A total of 94 patients were assigned to standard oxygen therapy, 106 to high-flow oxygen therapy, and 110 to noninvasive ventilation. The median interval between randomization and the initiation of treatment was 60 minutes (interquartile range, 11 to 120).Characteristics at InclusionThe characteristics of the patients at enrollment were similar in the three groups (Table 1Table 1Characteristics of the Patients at Baseline, According to Study Group.). The main cause of acute respiratory failure was community-acquired pneumonia, which was the diagnosis in 197 patients (64%). Bilateral pulmonary infiltrates were present in 244 patients (79%), and 238 patients (77%) had a PaO2:FIO2 of 200 mm Hg or less at the time of enrollment (Tables S1 and S3 in the Supplementary Appendix). The mean (SD) baseline FIO2, as measured through the nonrebreather face mask in 286 patients, was 0.650.13.TreatmentsThe initial mean settings were as follows: in the standard-oxygen group, an oxygen flow rate of 135 liters per minute; in the high-flowoxygen group, a gas flow rate of 4811 liters per minute, yielding a mean FIO2 of 0.820.21; and in the noninvasive-ventilation group, a pressure-support level of 83 cm of water, a PEEP of 51 cm of water, and an FIO2 of 0.670.24, resulting in a tidal volume of 9.23.0 ml per kilogram. Noninvasive ventilation was delivered for 8 hours (interquartile range, 4 to 12) on day 1 and for 8 hours (interquartile range, 4 to 13) on day 2.Primary and Secondary OutcomesThe intubation rate at day 28 was 38% in the high-flowoxygen group, 47% in the standard-oxygen group, and 50% in the noninvasive-ventilation group (P=0.18; P=0.17 by the log-rank test) (Figure 2AFigure 2KaplanMeier Plots of the Cumulative Incidence of Intubation from Randomization to Day 28.). The intervals between enrollment and intubation, as well as the reasons for intubation, did not differ significantly among the three groups (Table 2Table 2Primary and Secondary Outcomes, According to Study Group.).The crude in-ICU mortality and 90-day mortality differed significantly among the three groups (Table 2 and Figure 3Figure 3KaplanMeier Plot of the Probability of Survival from Randomization to Day 90.). The hazard ratio for death at 90 days was 2.01 (95% confidence interval [CI], 1.01 to 3.99) in the standard-oxygen group as compared with the high-flowoxygen group (P=0.046) and 2.50 (95% CI, 1.31 to 4.78) in the noninvasive-ventilation group as compared with the high-flowoxygen group (P=0.006; P=0.02 by the log-rank test) (Figure 3). The risk of death at 90 days remained significantly lower in the high-flowoxygen group after adjustment for the baseline Simplified Acute Physiology Score II and history of cardiac insufficiency (Table 2). Four patients died in the ICU without having undergone intubation (two in the standard-oxygen group and one in each of the other two groups). The 90-day mortality among patients who required intubation did not differ significantly among the groups (Table 2). The number of ventilator-free days at day 28 was significantly higher in the high-flowoxygen group than in the other two groups (Table 2).In a post hoc analysis, there was a significant interaction between the PaO2:FIO2 at enrollment (200 mm Hg vs. >200 mm Hg) and the treatment group with respect to status regarding intubation (P=0.01). In the subgroup of patients with a PaO2:FIO2 of 200 mm Hg or less, the intubation rate was significantly lower in the high-flowoxygen group than in the other two groups (Figure 2B and Table 2, and Table S4 in the Supplementary Appendix). The risk of intubation remained significantly lower in the high-flowoxygen group after adjustment for bilateral pulmonary infiltrates, respiratory rate, and preexisting history of cardiac insufficiency.The rate of various complications during the ICU stay did not differ significantly among the groups (Tables S2 and S4 in the Supplementary Appendix). Among the 40 patients who received noninvasive ventilation as rescue therapy, 19 of 26 patients (73%) in the standard-oxygen group and 9 of 14 (64%) in the high-flowoxygen group were intubated subsequently.Patient Comfort and SafetyAt 1 hour after enrollment, the intensity of respiratory discomfort in the patients was reduced and the dyspnea score was improved with the use of high-flow oxygen, as compared with the other two strategies of oxygenation (Table S5 in the Supplementary Appendix). There was no significant difference among the groups in the overall incidence of serious adverse events. Among the 18 episodes of cardiac arrest, 3 occurred before intubation (1 in the standard-oxygen group and 2 in the high-flowoxygen group). Two patients died during the process of intubation.DiscussionIn this multicenter, randomized, open-label trial, neither noninvasive ventilation nor high-flow oxygen decreased the rate of intubation (the primary outcome) among patients with acute hypoxemic respiratory failure. High-flow oxygen therapy, as compared with standard oxygen therapy or noninvasive ventilation, resulted in reduced mortality in the ICU and at 90 days.When planning the study, we assumed an intubation rate of 60% in the standard-oxygen group on the basis of data from previous randomized, controlled trials.7,9,10 Our results showed a lower rate than expected in the standard-oxygen group (47%) but also a higher rate than expected among patients treated with noninvasive ventilation (50%). The intubation rate in the noninvasive-ventilation group in our study is, however, consistent with the rates of 46 to 54% observed in other studies that included patients with acute hypoxemic respiratory failure.11-13,29 In a few observational studies,21,24,30 lower rates of intubation were seen among patients with hypoxemia who were receiving high-flow oxygen therapy than among those receiving noninvasive ventilation or standard oxygen therapy.The lower mortality observed in the high-flowoxygen group may have resulted from the cumulative effects of less intubation particularly in the patients with severe hypoxemia (PaO2:FIO2 200 mm Hg), as compared with other patients, and a slightly lower mortality among intubated patients who were treated with high-flow oxygen therapy than among those who were treated with one of the other strategies (Table 2). Two studies have also suggested that a failure of noninvasive ventilation might result in excess mortality, possibly because of delayed intubation,12,31 but we found no significant difference among the groups in terms of the time until intubation or the reasons for intubation. In our study, noninvasive ventilation that was administered to patients with severe lung injury could have increased the incidence of ventilator-induced lung injury by increasing tidal volumes that exceeded 9 ml per kilogram of predicted body weight.32-34 High-flow oxygen was also associated with an increased degree of comfort, a reduction in the severity of dyspnea, and a decreased respiratory rate. These findings might result from the heating and humidification of inspired gases, which prevented thick secretions and subsequent atelectasis but also from low levels of PEEP generated by a high gas flow rate16,17 and flushing of upper-airway dead space.20,21,23,24Our trial had several strengths that suggest that the results may be generalized to patients admitted for nonhypercapnic acute hypoxemic respiratory failure in other ICUs. These strengths included the multicenter design and sealed randomization to the assigned strategy, a well-defined study protocol that included prespecified criteria for intubation, complete follow-up at 90 days, and an intention-to-treat analysis.The main limitation of our study was the low power to detect a significant between-group difference in the intubation rate in the overall population. A reduced intubation rate was detected in the post hoc analysis in the subgroup of patients with a PaO2:FIO2 of 200 mm Hg or less, which was justified by a significant interaction between PaO2:FIO2 stratum and treatment.35In conclusion, treatment with high-flow oxygen improved the survival rate among patients with acute hypoxemic respiratory failure, even though no difference in the primary outcome (i.e., intubation rate) was observed with high-flow oxygen therapy, as compared with standard oxygen therapy or noninvasive ventilation.